The Quest for Quests: Part Seven (Enumeration 1)

You may want to start this series with part one, two, three, four, five or six.

What do we mean by failure in games? Let's start with a simple enumeration of the ways we can experience failure in a game.

The first is losing: a subset of all games - mostly those which are multi-player - are competitive and it is possible to win or lose such games (The rankings may be more sophisticated than the binary state, but assume for the moment that anything except coming first is a loss). Whether a player loses is defined by the rules of the game: this initially appears an important characteristic because most failures I'll be talking about are meta-game; that is fall outside of what would normally be considered part of the game.

But this lack of meta-game component is an incorrect assumption, because the game rules say nothing about the attributes of the players. Take a simple game, like rock, paper, scissors, where the rules are entirely proscribed and easily understood. In a single game of rock, paper, scissors against an opponent you have no other information about there is nothing outside game to influence your decision. But as soon as you play more than one game of rock, paper, scissors against the same opponent you have meta-game information: namely the history of your opponents decisions. It is entirely possible to gain significant advantage through analysis of these decisions - as well as any other information about the opponent - and there are a number of AI programming competitions for similar simple games which take advantage of this fact. In this instance, losing a single game can not be considered a failure, because the information derived from a loss can be used to improve the chance of successive wins.

Paradoxically, it is also possible to play two losing games alternately, and still end up winning overall (See this wikipedia article on Parrondo's paradox). So our intuition that losing in a game is equivalent to failure is incorrect for even simple games against another player.

Is it possible to lose against a computer opponent in the same sense that you can lose against another person? This is an important question, because if this is the case, it means we can lose in a single player game - the significance of which we'll look at later.

I'm going to use a simplified version of rock, paper, scissors to analyze this question: call it paper/shotgun. In this variant, we have two players, each of whom get to choose either paper or shotgun. If both make the same decision, player 1 wins - if the decision differs, player 2 wins. We'll encode paper as 0, and shotgun as 1, and then each player's decisions on successive games will resemble a binary string e.g. 001000111110...

Now if this binary string is perfectly random, there is no information that can be derived from the string - it's just noise. However, if the string is less than random - that is, there is information that can be derived because the string is compressible and an underlying set of rules is used to generate the string - then it is possible for one of the player's to derive the underlying system generating the other player's decisions and then start to win successive games. If player 1 is a set of rules, and player 2 is a perfect AI, then once the rules have been derived, player 2 will win consistently when it is possible to do so. If player 2 is less than a perfect AI - such as an actual human being, he may not be able to derive the rules perfectly, in which case successive plays can be used to refine the ruleset and improve the winning ratio.

Not there may still be random inputs to player 1's strategy, which prevent player 2 winning perfectly. But once player 2 has derived the underlying system of rules to player 1's choices, or has done so to the best of their capability, the game degenerates into application of the counter to the ruleset derived: it becomes a mechanical process.

This is the second failure in games: where player understanding of the game degenerates to a rote process which requires no analysis. Take tic, tac, toe. The entire game tree (the combination of all possible decisions) has been mapped and so any rational approach to playing the game is will result in a draw.

But people are not rational, which is why a game like tic, tac, toe still exists and is considered a game. You may choose to play it to explore the game tree and confirm that a draw is the logical outcome. You may choose to play against an opponent who is not aware of the correct decisions at each juncture of the tree, to prove your superior rationality, or to lose deliberately as a sop for their ego. On one level, people enjoy rote, mechanical processes because our brains are wired to enjoy a mix of novelty and mundane, as opposed to complete novelty the entire time - some pleasure in rote processes have enabled our ancestors to survive and evolve. The presence of grinding in so many games is testament to this.

Denigrating degenerate games and grinding is as much about game design snobbery as it is about the game itself. I dislike rule sets that result in degenerate game play, but I have to acknowledge that these games exist and are successful, and are an important part of many more sophisticated games. Organising your cards in Solium Infernum, clicklets in Bunni Game: How we First Met and kiting in Torchlit are all examples of mechanical processes which contribute to the feel of the game.

In the discussion so far, I have assumed that the player is capable of improving their ability to play the game as they play successive games or progress through successive decisions in a single game. That is not necessarily the case: and the third example of failure is failure to learn.

Failure to learn is about treating the game as a pedagogical instrument - a teaching device. In parts eight and nine, I'll be talking about how games teach, and more types of failure in games, and what other game designers and the wider world have been saying about failure.